In neotropical forests, 50–90% of the canopy trees bear fruits adapted for animal dispersal whereas close to 100% of the shrubs and sub-canopy trees produce fleshy fruits (Howe & Smallwood 1982). It has been suggested that during periods of fruit scarcity, some plant species perform a critical role in the forest ecosystem by sustaining frugivorous animals that are important dispersal agents for seeds of many other trees during other seasons of the year (Howe 1984, Terborgh 1986a). Loss or absence of those seed dispersers would have strong negative consequences for tree species, thereby affecting the health of the entire ecosystem, over time. For that reason, plant species that sustain frugivores (= seed dispersers) during periods of resource scarcity have been called ‘keystone species’ (Gilbert 1980, Howe 1984, Terborgh 1983, 1986a, b; van Schaik et al. 1993). For example, in the tropical rain forest of Manu National Park, Peru, Terborgh (1983) found that figs, palm nuts and nectar are resources of great importance for the primate community during periods of fruit scarcity. During that season, three species of fig were identified as keystone resources maintaining nearly 40% of the animal biomass in the ecosystem (Terborgh 1983).

In lowland tropical and temperate forests, nitrogen (N) and phosphorus (P) resorption from senesced leaves may reflect a mechanism of conservation of a limiting nutrient (Edwards & Grubb 1982, Killingbeck 1996, Proctor et al. 1989, Scott et al. 1992, Songwe et al. 1997, Vitousek & Sanford 1986). At the ecosystem level it has important implications for element cycling. The nutrients which are resorbed during leaf senescence are directly available for further plant growth, which makes a species less dependent on current nutrient uptake. Nutrients which are not resorbed, however, will be circulated through litterfall in the longer term (Aerts 1996).

The annual flood pulses of central Amazonian rivers inundate marginal forests to 10 m depth or more (Furch & Otto 1987, Junk et al. 1989). The regularity of this flooding seems to have promoted the development of a complex relationship between forest plants and fishes that consume their seeds (Goulding 1980). Several fish species migrate into the flooded forests to feed on buoyant fruits and seeds, and many plant species reach their fruiting peak during the flood season. Fruits and seeds have been reported to have adaptive features related to water (hydrochory) and/or fish (ichthyochory) dispersal (Araújo-Lima & Goulding 1998, Gottsberger 1978, Goulding 1980, Kubitzki & Ziburski 1994, Scarano 1998, Williamson 1999, Williamson & Costa 2000). Strong multicuspidate teeth in frugivorous fishes, fruit/seed buoyancy (Goulding 1980) and chemical signs released by fallen fruit to attract fish (Araújo-Lima & Goulding 1998) are mechanisms for the seed dispersal of the flooded forest community.

The distribution of the arboreal ant community plus a termite species of the genus Nasutitermes was inventoried on 938 red mangrove, Rhizophora mangle L., trees in a completely flooded mangrove forest of the Sian Ka'an Biosphere Reserve, Quintana Roo, Mexico. Colonies sheltered in dry hollow branches of the trees and the pseudobulbs of the epiphytic orchid, Myrmecophila christinae. Two size classes of dry hollow tree branches were categorized in order to test differences in ant distribution. As some trees remained unoccupied by either an ant or a termite colony, we deduced that the competition for nesting sites was low. Differences in the composition of the ant community corresponded to the nature of the shelters (i.e. diameter of the hollow branches or orchid pseudobulbs). The ant fauna was richer in the large dry hollow branches of R. mangle than in the smaller ones, with certain ant species belonging to the subfamilies Ponerinae and Formicinae being significantly more frequent in the large dry hollow branches than in the small ones. Cephalotes and Pseudomyrmex were the most frequent ant genera inhabiting the dry branches of R. mangle, while Dolichoderus bispinosus was the most frequent ant species inhabiting the orchid pseudobulbs. Arboreal Nasutitermes sp. selected mostly the orchid pseudobulbs and thus indirectly interfered with ant nest-site selection. Our results highlight niche differentiation through the selection of nest sites among different types of shelter.

This study demonstrates that elephant seed-dispersal is vital for Balanites wilsoniana (Zygophyllaceae), a forest canopy tree with no other effective dispersers. Non-dispersed seeds suffered high mortality (84%) and low germination (3%). Elephants were reliable at visiting fruiting trees (46%), ingesting available seeds (55%) and dispersing the fruit crop (26%). Compared with unpassed seeds, seeds passed through an elephant's gut had improved germination (54.9% vs. 2.9%) and reduced time to germination (82 vs. 132 d). Given predation rates under parent trees, elephant gut passage is expected to improve absolute germination by 66%. When this change in germination is divided by germination of unpassed seeds, elephant gut passage is expect to improve relative germination by as much as 4000%. Although post-dispersal predation occurred in dung containing B. wilsoniana seeds, dispersed seedlings were abundant. Compared with seedlings under parent trees, dispersed seedlings had greater survival and height. Dispersed seedlings survived in a diversity of environments, thrived under high-light conditions, and were negligibly affected by post-dispersal density. Although elephants are not essential for B. wilsoniana germination, their actions greatly increase seedling establishment and move a large proportion of seeds. This high degree of effectiveness, together with very low non-dispersed seed survival, provides strong evidence that B. wilsoniana is dependent on elephants for its long-term persistence.

Vertebrate seed dispersers could impact the evolution of seed size or alter the pattern of seedling recruitment if they responded differently to seeds of varying size (Jordano 1995). For example, models of seed caching by birds and mammals predict that seeds of higher nutritive value should be placed farther from parent trees and in lower densities than lower quality seeds (Clarkson et al. 1986, Stapanian & Smith 1978, Tamura et al. 1999). Comparisons of seed removal rates among tropical tree species in South-East Asia (Blate et al. 1998) and Australia (Osunkoya 1994) failed to show a relationship between seed size and removal rate, although the probability that a seed was scatterhoarded by agoutis (Dasyprocta punctata) in a neotropical rain forest increased with interspecific seed size (Forget et al. 1998).

The Brazilian Atlantic Forest is one of the world's ‘hot spots’ for conservation because of its high level of endemism and number of endangered species (Myers et al. 2000). After centuries of deforestation, most of the remaining Atlantic Forest is scattered as small fragments on private land (Fonseca 1985, Jorge & Garcia 1997). Pollination could be impaired in these fragments because isolation often negatively influences the dispersal of insects, the most important pollinating agents in the tropics.

Interest in the colour of fruits eaten by vertebrates has a long history (Ridley 1930, Turcek 1963, van der Pijl 1972). Within a given geographical area, colours of bird-consumed fruits compared with mammal-consumed fruits are fairly consistent (Janson 1983, Wheelwright & Janson 1985, Willson & Whelan 1990), but comparisons among different regions reveal selective pressures of different animal groups (Lord 2002, Willson et al. 1989). One consistent pattern among regions is that fruits with a two-colour display typically are eaten by birds (Wheelwright & Janson 1985, Willson & Thompson 1982, Willson & Whelan 1990). Thus, it is thought that such bicoloured displays are specifically adapted to attract birds and plants with these displays presumably benefit from seed dispersal by birds (Janson 1983, Willson & Whelan 1990).

Ranging patterns are important features of animal ecology both for evolutionary ecologists who study adaptive foraging and habitat use, and for conservationists who need information on habitat requirements and minimum resource needs. Because ranging patterns are difficult and time-consuming to measure, many researchers have sought socio-ecological variables that predict range use. For example, many studies show a correlation between daily path length (day-range length) and the number of animals in a group (Chapman et al. 1995, van Schaik et al. 1983, Waser 1977, Wrangham et al. 1993). This seems reasonable because larger groups have more mouths to feed, necessitating more travel to find adequate food supplies. Group size is far easier to measure than day range, hence the attraction of this variable. However, group size must be used cautiously as a proxy for day range because their relationship is neither strong nor straightforward.

Large rivers have played a prominent role in biogeographic theory for their potential to act as barriers for the dispersal of terrestrial organisms,and therefore be involved in the generation of species diversity (Brown & Lomolino 1998). In this paper, we document the potential role of macrophyte rafts as a mechanism by which Amazonian rivers could act as dispersal agents rather than barriers, transferring organisms across banks and possibly across very large distances. These vectors could therefore act against speciation and towards homogenization of the local biota.

Tropical rain forests have a high richness of plant species. However, the richness of insect species associated with the forest canopy is still poorly known. Estimates of arthropod fauna have followed several approaches from surveys on a single or few trees (Erwin 1988,1990) to regional sampling of specific insect guilds such as gall-inducing insects (Fernandes & Price 1988, Lara & Fernandes 1996, Price et al. 1998).

As part of the programme Écosystèmes et Paléoécosystémes des Forêts Intertropicales (ECOFIT,France),we are engaged in multisite comparative studies of how forests regenerate across ecotones such as those between forests and savannas (Vincens et al. 2000).One such ecotone comprises the forest succession from bare soil that occurs naturally on beach successions along meandering rivers. To gauge the development of foliar layers during succession and quantify changes of efficiency of the vegetation in capturing the available solar energy, we are measuring the Leaf Area Index (LAI) of transects within forests and across their edges.

Data on the pollination biology of neotropical aroids are still scarce and concern only a few species (Beath 1999, Croat 1997, Gibernau et al. 1999, 2000; Mayo et al. 1997 and references cited therein). It appears from these studies that Anthurium, Monstera and Spathiphyllum are on the whole pollinated mainly by bees (but see Kraemer & Schmitt 1999), whereas Dieffenbachia, Homalomena, Syngonium, Philodendron and Xanthosoma are generally beetle-pollinated. Although araceous inflorescences may be visited by several insect taxa (Madison 1979), only a few are the legitimate pollinators for each species (Seres & Ramirez 1995, Valerio 1984, Young 1986).

Most researchers assume litter decomposition to take place at the site where litterfall occurs, mostly by soil micro-organisms and microfauna of less than 1 mm in size. Accordingly, the litterbag method has become the standard technique for evaluating litter decomposition rates (Schlesinger 1985, Wedderburn & Carter 1999). However, there is evidence indicating that at least in tropical arid ecosystems litter can be removed from the site in substantial amounts by animals, including large vertebrates. For example, litter is removed for forage or as building material for nests by ants (particularly leaf-cutting ants) and termites (Bucher 1982, Scholes & Walker 1993). Moreover, ruminant species (including domestic cattle and goats) supplement their food intake by feeding on falling leaf litter during the dry season in African savannas (Owen-Smith & Cooper 1987) and in the Chaco savannas of Argentina (Morello & Saravia-Toledo 1959). These observations suggest that litter removal may be an important component of the decomposition process, particularly in tropical semi-arid environments where lack of soil moisture decreases microbial activity, increasing litter availability for herbivores.

Fine roots are very important in ecology because of their role in nutrient and water uptake, and as a source of organic matter to the soil. In carbon-cycle models, fine roots are a significant organic matter pool (Potter 1999) with high net primary productivity (Jackson et al. 1997) and turnover (Gill & Jackson 2000). In ecosystem production studies, fine roots are usually divided between biomass and necromass.The sum of both pools is defined as root mass (Klinge & Herrera 1978) or total root biomass (Böhm 1979). In tropical forests the study of fine-root biomass is restricted because of the difficulties in distinguishing live roots. Visual methodologies are not adequate in tropical forests where high diversity is expressed through many root morphologies. On the other hand, definitions of root death are ambiguous and differ between different studies (Comas et al. 2000). Fine-root mass is easier and more accurate to measure than fine-root biomass because subjective selection criteria are avoided. However, in the measurement of below-ground production, the estimation of fine-root biomass is essential (Jackson et al. 1997). The adaptation of objective selection methods (Comas et al. 2000, Joslin & Henderson 1984) to measure live and dead root fractions is urgently needed.

Habitat fragmentation affects the ecology of tropical rain forests in many ways, such as reducing species diversity of many taxa (Laurance et al. 2002, Lovejoy et al. 1986) and increasing rates of tree mortality and canopy-gap formation near forest edges (Laurance et al. 1997, 1998, 2001). Such obvious alterations have been documented in many fragmented forests, but more subtle changes, such as those affecting plant phenology (the timing and frequency of flower, fruit and leaf production), have received far less attention. Adler & Kiepinski (2000) showed that different populations of the successional tree Spondias mombin on small man-made islands in Panama had highly synchronous flowering and fruiting. In montane forests in Colombia, Restrepo et al. (1999) demonstrated that under-storey fruit abundance was consistently increased over time near forest edges relative to forest interiors. Beyond these and a few other studies (Ackerly et al. 1990, Nason & Hamrick 1997), however, the effects of fragmentation on plant phenology have been inadequately assessed, especially in the tropics.

VOLUME 19, NUMBER 1, JANUARY 2003, PAGES 19-25

Variation in local abundance of herbivores is influenced by several abiotic and biotic factors, but it may often be determined by the local availability of plant resources (Quinn et al. 1997). Leaf-cutting ants, common defoliators of plants in the Neotropics, are more abundant in early successional forests than in old-growth forest (Jaffe & Vilela 1989, Vasconcelos & Cherrett 1995). A recent study showed that the high density of mature leaf-cutting ant nests in early successional forests was mainly determined by the availability of pioneer plant species, largely preferred by the leafcutters (Farji-Brener 2001). The dominance of pioneer species in early successional forests decreases foraging costs for leaf-cutting ant colonies, and thus these habitats can support a high density of mature colonies. On the other hand, the effective defence mechanisms of shade-tolerant species, which dominate primary habitats, could explain the low density of mature leaf-cutting ant colonies in old-growth forests (palatable forage hypothesis, see Farji-Brener 2001). According to this hypothesis, it is proposed that leaf-cutting ant colonies in old-growth forests should concentrate their foraging effort in treefall gaps, in which sun and pioneer plant species (preferred by leafcutters), are readily available (Farji-Brener 2001, Nichols-Orians 1991a, b; Wetterer 1994). However, this prediction of the palatable forage hypothesis has not yet been explicitly documented.

An incomplete version of Figure 2 was published in the previous issue. The correct version is shown below. The publishers would like to apologise to the authors and their readers for this error.